Neuron Explorer

Qwen3-0.6B

"in" right before a year

A neuron in layer 7 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

Fires on the word "in" just before a year number, such as "died in" or "passed Congress in", so the model expects a date next. Turning it down made the text repeat itself; turning it up gave no clear change.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).

Fires most on

The text windows where this neuron's activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    948 election to hard-line nationalists who institutionalised apartheid. He continued to work for reconciliation and emphasised the British Commonwealth's positive role until his death in 19

  2. #2 peak

    the UAR would be a combination of the then-Egyptian anthem and "Homat elDiyar" When Syria seceded from the union in 1961, the⏎anthem was completely restored.⏎Translation of Syrian

  3. #3 peak

    hips, also published a reflection on Religious Freedom Day.⏎Rogers celebrates the Religious Land Use and Institutionalized Persons Act, which passed Congress by unanimous consent in 2000 with backing from a politically and religiously diverse coalition. RL

  4. #4 peak

    980s, it wasn’t until the LIGO detector came online in 2002 that the hunt for elusive spacetime ripples started to get

  5. #5 peak

    – which looks to celebrate women’s achievement, raise awareness against gender bias and take action for equality. The first IWD was acknowledged in the U.S. in 1909 when 15,000 female workers went on

  6. #6 peak

    face of the pandemic of obesity, overweight, and diabetes that lives in the country, should be changed.”⏎The Mexican government declared obesity to be a national epidemic in 2016.⏎INSP said it wanted to make clear its position in

Steering

Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.

turned down a lot
Prompt

Here is something I have been thinking about lately: that is, the things that are going to come to me. So, I am doing something. I am doing something in my own life. So, the things that are coming to me, so I am going to be in my own

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    vascular Diagnostic Center, APMC, in West Monroe, LA, angioplasties are performed by our cardiologist, Dr. Virginia Gonzalez.⏎Purpose of Angioplasty⏎The coronary arteries are the arteries that carry blood to the heart. Unfortunately

  2. #2 peak

    , but the stent remains in place. Patients typically stay overnight after undergoing an angioplasty through our cardiology practice in West Monroe, LA.⏎An angioplasty can improve your heart health and reduce the risk of heart attack by opening up

  3. #3 peak

    As the balloon inflates, it opens up and expands the artery by pushing away plaque and cholesterol build-up. The balloon might be inflated several times during the angioplasty to fully open up the artery so enough blood can flow to the heart.

  4. #4 peak

    purpose of an angioplasty is to clear plaque and cholesterol build up in the coronary arteries to improve blood flow to the heart.⏎The procedure for performing an angioplasty involves the use of a catheter and an expandable balloon to open up

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L5:70 excites
  2. L6:951 excites
  3. L6:764 excites
  4. L6:1954 suppresses
  5. L3:1153 suppresses
  6. L5:2445 suppresses
  7. L6:415 weakly suppresses
  8. L3:2153 weakly suppresses
  9. L6:964 weakly excites
  10. L5:218 weakly excites
  11. L2:841 weakly suppresses
  12. L6:954 weakly suppresses
  13. L6:1946 weakly excites
  14. L5:237 weakly suppresses
  15. L5:2741 weakly suppresses
  16. L6:2301 weakly excites
  17. L3:2345 weakly excites
  18. L6:657 weakly excites
  19. L4:1881 weakly excites
  20. L4:2349 weakly excites
  21. L6:1443 weakly excites
  22. L5:2759 weakly suppresses
  23. L5:895 weakly excites 'well' after is or are
  24. L4:178 weakly suppresses

Feeds later layers

  1. L10:915 suppresses
  2. L8:1880 excites
  3. L19:1348 excites
  4. L10:1925 suppresses
  5. L10:2361 excites
  6. L8:857 excites
  7. L11:834 suppresses
  8. L18:1378 suppresses
  9. L14:2297 excites
  10. L12:599 excites
  11. L22:1063 excites
  12. L9:2250 excites
  13. L11:1679 excites
  14. L9:447 suppresses
  15. L9:1725 excites
  16. L10:1690 excites
  17. L9:221 excites
  18. L12:1183 suppresses
  19. L9:1674 excites
  20. L12:2917 suppresses
  21. L9:2831 suppresses
  22. L10:1010 weakly excites 'in the process of' and continuing actions
  23. L21:398 weakly suppresses
  24. L17:1416 weakly suppresses

Activation histogram

How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.

1 values between -4.69 and -4.481 values between -2.63 and -2.431 values between -2.43 and -2.221 values between -1.61 and -1.403 values between -1.40 and -1.194 values between -1.19 and -0.9915 values between -0.99 and -0.7826 values between -0.78 and -0.5854 values between -0.58 and -0.37203 values between -0.37 and -0.171967 values between -0.17 and 0.041231 values between 0.04 and 0.25281 values between 0.25 and 0.4563 values between 0.45 and 0.6620 values between 0.66 and 0.8611 values between 0.86 and 1.077 values between 1.07 and 1.274 values between 1.27 and 1.486 values between 1.48 and 1.681 values between 1.68 and 1.892 values between 1.89 and 2.101 values between 2.10 and 2.303 values between 2.51 and 2.712 values between 3.33 and 3.53offnegativepositiveoften

Logit lens

The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layers instead.

Promotes

  1. ngoing
  2. àng
  3. orang
  4. 看着
  5. anmar
  6. cache
  7. 不断发展
  8. ọng
  9. .getPage
  10. 统计局

Suppresses

  1. 两次
  2. Inserted
  3. �
  4. toList
  5. 适当
  6. insert
  7. 一次
  8. Inserts
  9. serter
  10. .lesson